swarm-review

You are in Swarm Suite navigator mode. The user typed /swarm-review $ARGUMENTS. The $ARGUMENTS (if any) is the scope path the user wants reviewed; if empty, default to the project root.

Do this:

1. Call kb_navigator_state(project_path=<cwd>) to learn pipeline state.
2. Resolve scope:
   • If $ARGUMENTS is non-empty: that's the scope.
   • Else: project root (current working directory).
3. Pipeline check:
   • If no pipeline: kb_start_pipeline(project_path=...).
   • If pipeline exists but current_stage != "review": tell the user the current stage and ask whether to advance/skip to review or work in-place.
4. Pre-pick experts: call kb_route_experts(task="review of <scope>", experts_dir=<path to review-swarm experts>, top_k=5, min_score=0.05) and tell the user which experts you're going to use + WHY (one line per expert).
5. Confirm with the user before running the actual review (it's expensive: ~5 LLM calls per file across selected experts).
6. On confirmation: call orchestrate_review(project_path=..., scope=<scope>, max_experts=<count>).
7. After Phase 1 + Phase 2 complete: call get_summary(session_id) and present the findings grouped by severity (HIGH first), with confirmed/disputed annotations.
8. Ask the user how they want to triage (approve all confirmed, dismiss all low-severity, etc.) and apply via bulk_update_status or per-finding mark_fixed.
9. End by calling kb_navigator_state again and offering the next step (typically: advance to fix stage).

If the user supplied a scope that doesn't exist, ask for clarification (one question only).